Socratic Learning

Why Socratic Learning?

Most AI platforms treat learning as an afterthought. Socratic Learning makes continuous improvement built-in:

• Interaction Tracking - Capture and store all agent interactions with full context for analysis
• Pattern Detection - Automatically identify recurring patterns in agent behaviors and model outputs
• Performance Monitoring - Track success rates, response times, and costs across all interactions
• Data-Driven Recommendations - Get actionable improvement suggestions based on detected patterns
• Fine-Tuning Ready - Export interaction data in industry-standard formats for model fine-tuning

A continuous learning system for AI agents that tracks interactions, detects patterns, and provides data-driven improvement recommendations.

Features

• Interaction Tracking - Capture and store all agent interactions with context
• Pattern Detection - Identify recurring patterns in agent behaviors and LLM outputs
• Performance Metrics - Monitor agent effectiveness with success rates, response times, costs
• User Feedback Integration - Collect and analyze user feedback on agent responses
• Learning Recommendations - Generate actionable improvement suggestions
• Fine-Tuning Export - Export interaction data for model fine-tuning
• Analytics & Reporting - JSON-based insights and metrics
• Framework Integration - Works with Openclaw and LangChain

Installation

# Core package
pip install socratic-learning

# With Socratic Agents integration
pip install socratic-learning[agents]

# With all optional dependencies
pip install socratic-learning[all]

# For development
pip install socratic-learning[dev]

Quick Start

from socratic_learning import LearningManager
from socratic_agents import SocraticCounselor

# Initialize learning manager
learning = LearningManager(storage="sqlite", db_path="learning.db")

# Create a tracking session
session_id = learning.create_session(
    user_id="user123",
    context={"environment": "production"}
)

# Track agent interactions
counselor = SocraticCounselor()
result = counselor.guide("recursion", level="beginner")

learning.track_interaction(
    session_id=session_id,
    agent_name="SocraticCounselor",
    input_data={"topic": "recursion", "level": "beginner"},
    output_data=result,
    model_name="claude-opus-4",
    provider="anthropic",
    input_tokens=150,
    output_tokens=500,
    duration_ms=1200.0,
)

# Add user feedback
learning.add_feedback(
    interaction_id=interaction.interaction_id,
    rating=5,
    feedback="Very helpful explanation!"
)

# Get metrics
metrics = learning.get_metrics(agent_name="SocraticCounselor")
print(f"Success rate: {metrics.success_rate}%")
print(f"Avg rating: {metrics.avg_rating}/5")

# Detect patterns
patterns = learning.detect_patterns(agent_name="SocraticCounselor")
for pattern in patterns:
    print(f"Pattern: {pattern.name} (confidence: {pattern.confidence})")

# Get recommendations
recommendations = learning.get_recommendations(agent_name="SocraticCounselor")
for rec in recommendations:
    print(f"Recommendation: {rec.title}")

# Export for fine-tuning
learning.export_for_finetuning(
    output_path="finetuning_data.jsonl",
    agent_name="SocraticCounselor",
    min_rating=4,
    format="openai"
)

Core Concepts

Interaction

Represents a single agent interaction with input, output, performance metrics, and optional user feedback.

Pattern

A detected recurring pattern in agent behaviors (e.g., error patterns, topic-specific behaviors).

Metric

Aggregated performance metrics (success rate, average response time, user satisfaction, costs).

Recommendation

An actionable improvement suggestion based on detected patterns and metrics.

Architecture

• Core Models - Dataclass-based models with serialization
• Storage Layer - Abstract interface with SQLite backend
• Tracking - Interaction logger with session management
• Analytics - Pattern detection and metrics collection
• Integrations - Openclaw skills and LangChain tools

Documentation

• Maturity Calculation System - Complete guide to the confidence-weighted maturity scoring system, including core algorithms, category definitions, and API reference
• See examples/ for complete working examples

Testing

# Run all tests
pytest

# Run with coverage
pytest --cov=src/socratic_learning --cov-report=html

# Run specific test file
pytest tests/unit/test_models.py -v

Code Quality

# Format with Black
black src/ tests/

# Lint with Ruff
ruff check src/ tests/

# Type check with MyPy
mypy src/

License

MIT

Contributing

Contributions welcome! Please open an issue or submit a PR.

Support Development

If you find this package useful, consider supporting development:

• Become a Sponsor - Get early access to new features
• Star on GitHub - Shows your appreciation
• Report Issues - Help improve the package

Your support helps fund development of the entire Socratic ecosystem.

Status

Phase 1 - Core foundation complete (v0.1.0 development)

• ✅ Core data models
• ✅ SQLite storage
• ✅ Unit tests
• 🚀 Phase 2-6 planned

Part of Socrates AI Ecosystem

This package is a component of Socrates AI, a production-ready platform for building intelligent multi-agent systems with constitutional governance.

Use This Package Standalone:

pip install socratic-learning

Or As Part of Socrates Platform:

pip install socrates-ai  # Includes 37+ modules + all 11 packages

Integration Example:

See the Socrates ECOSYSTEM.md for detailed integration examples showing how to use socratic-learning with other Socratic packages.

Related packages you might use together:

• See Complete Package Map

More Information:

• 📖 Full Socrates Documentation
• 🏗️ Complete Architecture Guide
• 💬 Socrates Discussions